community update, Thursday, January 10, 2008
Michael Shiloh
michael at openmoko.org
Fri Jan 11 00:43:34 CET 2008
Hi Ian,
ian douglas wrote:
>> Although it would be nice to take advantage of this faster processor, 3
> > problems make this impractical
>
> Michael,
>
> If I'm reading your message correctly, the GTA02 will indeed have a
> 500MHz processor, but will not be running at 500MHz because of the three
> problems you outlined, is that correct?
That is correct. We are installing a CPU that is rated for a maximum
speed of 500MHz, but clocking it at 400MHz.
This is fairly common in the hardware world as availability of certain
chips changes.
For example, a manufacturer might find they can make the faster chips
for the same price as the slower, and it costs them more to stock the
two speeds, so they produce only the faster.
Another reason might be that the volume of sales of the 400MHz part was
so low that it wasn't worth manufacturing.
>
> Does this mean you will be slowing the CPU to 400MHz, or some other
> speed, or will you only know the answer once more testing has been done?
I'm not sure the expression "slowing the cpu" is accurate. The CPU is
rated for a maximum clock frequency of 500MHz, but it can be used at
other speeds as well.
If you are asking whether we intend to increase the clock speed at a
later time, the answer is it's possible, but it's not planned. Since the
rest of the circuit is designed for 400MHz it would take some serious
analysis and testing to convince ourselves that it works reliably at 500MHz.
A big part of hardware design consists of making sure that all signals,
taking into account worst-case propagation delays and rise and fall
times, will arrive at their destinations early enough to meet the setup
times of the destination. This analysis is done at the intended CPU
clocking frequency, which in our case was 400MHz. There is no
expectation that these conditions will be met when the CPU is clocked at
500MHz - rather, every single signal in the circuit must be analyzed at
this higher frequency.
This is a tremendous amount of work, and is properly prioritized below
getting GTA02 into manufacturing.
>
> Your description of problem 2 makes it sound like memory access will be
> slower, at 83MHz, if running the CPU at full speed because of the memory
> bus speed, instead of the anticipated 100MHz, is that correct?
I can't answer for sure because I didn't work this out myself - rather I
asked someone else. I can only presume that dividing by 5 was not an
option. I'll find out.
This frequency pre-scaler is not part of our circuitry; rather, it's
built-in to the system-on-a-chip (SoC). Typically, those system provide
pre-scalers, and typically those pre-scalers provide a limited range of
fixed numbers for dividing the incoming frequency. So just because "4"
and "6" were available divisors does not indicate that "5" is available
as well.
> Does
> anyone there anticipate that this 17% decrease in memory access speed
> will be noticeable to the end user? What's the memory access speed on
> the GTA01?
There is no decrease in memory speed - the system was designed to run
with a CPU clock of 400MHz, and the memory at 100MHz, and that's what it
will do.
>
> Thanks,
> Ian
Thanks,
Michael
More information about the community
mailing list